XRL: An FMM-Accelerated SIE Simulator for Resistance and Inductance Extraction of Complicated 3-D Geometries

Mingyu Wang,Ping Liu,Jihong Gu,Xiaofan Jia,Abdulkadir C. Yucel

2024-09-19

Abstract:A fast multipole method (FMM)-accelerated surface integral equation (SIE) simulator, called XRL, is proposed for broadband resistance/inductance (RL) extraction under the magneto-quasi-static assumption. The proposed XRL has three key attributes that make it highly efficient and accurate for broadband RL extraction of complicated 3-D geometries: (i) The XRL leverages a novel centroid-midpoint basis transformation while discretizing surface currents, which allows converting edge-based vector potential computations to panel-based scalar potential computations. Such conversion makes the implementation of FMM straightforward and allows for drastically reducing the memory and computational time requirements of the simulator. (ii) The XRL employs a highly accurate equivalent surface impedance model that allows extracting RL parameters at low frequencies very accurately. (iii) The XRL makes use of a novel preconditioner, effectively including both diagonal entries and some near-field entries of the system matrix; such preconditioner significantly accelerates the iterative solution of SIE. The proposed XRL can accurately compute broadband RL parameters of arbitrarily shaped and large-scale structures on a desktop computer. It has been applied to RL parameter extraction of various practical structures, including two parallel square coils, a ball grid array (BGA) package and a high brand package on package. Its application to the parameter extraction of the BGA shows that the XRL requires 93.2x and 14.2x less computational time and memory resources compared to the commercial simulator Ansys Q3D for the same level of accuracy, respectively.

Signal Processing

What problem does this paper attempt to address?

The problem that this paper attempts to solve is to develop an efficient and accurate broadband resistance/inductance (RL) extraction method for complex three - dimensional geometries. Specifically, the paper proposes an XRL, a surface integral equation (SIE) simulator accelerated by the fast multipole method (FMM), to perform broadband RL parameter extraction under the magneto - quasistatic assumption.

### Core of the Problem
1. **Limitations of Existing Methods**:
- **Volume Integral Equation (VIE) Simulators**: Such as FastHenry and TetraHenry, etc. Although they can handle complex 3D geometries, they require a large number of volume elements at high frequencies, resulting in low computational efficiency.
- **Existing SIE Simulators**: Although more efficient than VIE simulators, they have insufficient accuracy in the low - frequency region or are not highly efficient when dealing with complex structures.
- **Commercial Simulation Software (such as Ansys Q3D)**: Although it has high accuracy in the low - frequency region, it uses an approximate model in the transition - frequency region, leading to inaccuracy.

2. **Requirements and Challenges**:
- There is a need for an SIE simulator that can maintain high accuracy and high efficiency in the range from very low frequencies to high frequencies.
- It can handle complex 3D structures of arbitrary shapes and large - scale.
- It can run on a desktop computer with low memory and computational time requirements.

### Main Contributions of XRL
To meet the above challenges, XRL proposes the following three key techniques:

1. **New Centroid - Midpoint (CM) Basis Function Transformation**:
- It converts the edge - based vector potential calculation into the panel - based scalar potential calculation, simplifies the implementation of FMM, and significantly reduces the memory and computational time requirements.

2. **High - Precision Equivalent Surface Impedance (ESI) Model**:
- It ensures the accuracy of RL parameter extraction in the low - frequency region and avoids the problem of requiring different solvers and approximate models in commercial software.

3. **New Pre - conditioner**:
- It combines diagonal terms and partial near - field terms, significantly accelerating the iterative solution process and reducing the number of iterations.

Through these improvements, XRL can efficiently and accurately calculate the broadband RL parameters of complex 3D structures on a desktop computer. Compared with the commercial software Ansys Q3D, it shows higher efficiency and lower resource requirements in multiple practical application cases. For example, when analyzing BGA packages, the computational time and memory resources required by XRL are 93.2 times and 14.2 times less than those of Ansys Q3D respectively.

### Summary
The paper aims to fill the gap in existing technologies in broadband RL parameter extraction and provide an efficient and accurate SIE simulator XRL suitable for complex 3D structures.